Monologue: Amazon Is Warehousing Billions In GPUs | Better Offline

In progressEd Zitron (Better Offline)podcast2026-10-23read planted ai llmsbusiness

Synopsis — AI-drafted from Dan's notes

Zitron’s weekly monologue takes up Business Insider’s 20 October report: under an “Early Rack Receipt” process, AWS expected about $5 billion of Nvidia GB300 racks in August and roughly $3 billion a month after, some of it parked in intermediate warehouses because sites lacked the power to switch it on. He reads it as confirmation of his own estimate of GPUs bought and not installed. Amazon used to pay when the gear reached the data centre and now pays earlier, and he sees Nvidia behind the change: channel stuffing, he calls it, buyers made to take delivery months before they can use it because they fear losing their place in line. If the chips can’t be plugged in, he asks, what makes them scarce? Amazon calls staging near the sites best practice and denies lacking a plan to deploy them.

He goes further, and this is his inference, not the report’s: buying whole servers months early soaked up memory and storage, so the jump in RAM and consumer electronics prices was, in his view, a shortage made to serve Nvidia’s quarterly growth. Most of the eleven minutes is anger at Jensen Huang, at hyperscaler chiefs and at anyone who defends them. He ends with a reckoning he expects after the bubble bursts, when people will be asked how gullible they were: who repeated the labs’ run-rate figures, who waved away the costs. LLMs may survive in some open form, he allows, but not the industry as it stands.

Connections